Add cross-platform memory map abstraction and use it in libfuzzer.

[gcoakes]

This is a collection of commits working towards both #20987 and #20803. They implement the following:

• Windows extern definitions for CreateFileMapping and MapViewOfFile.
• A cross-platform abstraction for mapping files into virtual memory.
• Using that abstraction in libfuzzer's seen_pcs field.

There is still work that needs to happen for both of those issues.

[gcoakes]

Note for reviewers, I removed MemoryMappedList from libfuzzer because I realized the current usage never grows the file. MemoryMappedList seems like it was added in preparation for using the same type to represent both fixed-size, memory-mapped files and dynamic-sized, memory-mapped files.

I think that is better handled by two separate types: (1) std.fs.MemoryMap and (2) std.fs.MemoryMapGrowable (name is flexible). I had both implemented in my second commit, but I realized std.fs.MemoryMapGrowable would not actually be used until working on memory-mapping the Fuzzer.input field (ziglang/zig/issues#20803). So, I stashed those changes and should have another PR implementing that.

[gcoakes]

Checks are failing for Linux due to non-native operating systems: WASI and Windows. I've been testing with zig build test-std -Dskip-non-native -Dskip-release as was recommended in the wiki.

WASI is a simple matter of producing a compile error when it is attempted to be used while targeting WASI. WebAssembly/WASI#304

Windows seems to be a Wine specific issue since it passes on my Windows development machine and CI reports success on Windows. I need to get Wine testing working on my Linux development machine.

Moving to draft until I can resolve those. I'll fork this branch to a non-PR one so that I can not waste your CI resources while working on that.

[gcoakes]

CI failed for aarch64 MacOS Release only (Debug passed). The error message was saying something about a missing feature. That makes me think it is unrelated to my changes.

This branch was several commits behind master. I rebased onto the latest master in the hopes of resolving the CI issue. In any case, I do not have access to a MacOS aarch64 machine to test it locally.

[gcoakes]

The CI failures appear spurious and unrelated to the changes here. This time it is on a different OS panicking on a debug assertion in a totally unrelated part of code:

error: thread 790374 panic: reached unreachable code
/home/ci/actions-runner11/_work/zig/zig/lib/std/debug.zig:399:14: 0x4ce6fe7 in assert (zig)
    if (!ok) unreachable; // assertion failure
             ^
/home/ci/actions-runner11/_work/zig/zig/lib/std/array_hash_map.zig:1959:11: 0x4ebd943 in capacityIndexType (zig)
    assert(bit_index <= 32);

I don't see how this patchset could even cause a regression inside the compiler since the newly added standard library APIs are not used in the compiler.